Discussion of Results

 

The quality of the reproduction of instruments was fine as a first order approach.  The clarinet sound certainly sounds similar to an actual clarinet and the flute sound also is a decent portrayal of an actual flute (though not quite as good as the clarinet reproduction).  If we had more time, we could have attempted to model the transients of the instruments (attack, release, decay).  This approach would have yielded even better representations of the instruments.

 

As can be seen in the graph for the morphed signal in time domain, we tried to morph these two signals by linearly increasing additive synthesis, meaning that we gradually insert the harmonic frequencies of the signal being morphed into, and increase the amplitude of the newly inserted harmonic series.  On the graph, an attempt of the first signal to merge into the second can be seen. However, our current approach could not map the amplitude of one signal to the other. As a result, at the end of morphing, there is still a sudden change to the next signal. 

 

We are thinking that in order to make a smoother transistion, or morphing, we can try cross-synthesis. Cross synthesis is also a type of spectral morphing. It involves impressing the spectral envelope of one sound onto the flattened spectrum of another envelope.

 

Specific steps for cross synthesis involves: 1) short-time Fourier Transform of signals, 2) use appropriate calculation to obtain appropriate spectral envelop, 3) use envelop to flatten the spectrum of the carrier and 4) obtain the morphing by multiplying envelop of modulator by the corresponding flattened spectral frame.

 

Cross synthesis should yield a better morphing between the two music signal that we have here. We are going to look closely at implementing cross synthesis and reseraching other more elegant approaches to audio spectral morphing in the future.